The short version
When someone asks an AI search tool “What's the best email marketing tool for a small shop?”, the answer is not a private opinion stored inside the model. For tools that search the web, the answer is usually built in three steps.
- 1
Search
The tool turns the question into one or more web searches.
- 2
Read a short list
It reads a handful of the pages those searches return: often review sites, “best of” roundups, directories and forum threads.
- 3
Answer
It writes an answer that repeats the brands those pages name, and often links to some of them as sources.
So the real question is not “Does the AI like us?” It is “Are we on the pages the AI reads for our buyers' questions?” The rest of this guide explains each step.
Step 1: the AI searches first
AI models are trained on old data, so for questions about products, prices and “best” lists, many AI search tools look things up on the live web before they answer.
Google says this openly. Its guide to AI features (opens in a new tab) explains that AI Overviews and AI Mode may use query fan-out: running several related searches across subtopics before writing one answer. The same guide says a page must be indexed and eligible to show in Google Search with a snippet to be shown as a supporting link.
OpenAI's crawler documentation (opens in a new tab) says sites that block its OAI-SearchBot crawler will not be shown in ChatGPT search answers. In other words, if the search step can't reach a page, the answer can't use it.
Step 2: it reads a short list of pages
The model doesn't read the whole web. It reads what the search step handed it: a small set of pages. Whatever those pages say becomes the raw material for the answer.
Think about what a search for “best email marketing tools” returns. Mostly pages whose whole job is to compare several tools. That's why these kinds of pages matter so much:
| Page type | Example (fictional brand Acme Email) | Why AI leans on it |
|---|---|---|
| Review profile | Acme Email's listing on a software review site | Real user ratings, side-by-side comparisons |
| Roundup article | “The 8 best email marketing apps” on a tech blog | Already answers the exact “best X” question |
| Directory or category page | An “Email marketing software” category with filters | Lists many options in one place |
| Community thread | A forum post: “What do you use for newsletters?” | Real people sharing first-hand experience |
| Brand's own site | acmeemail.com | Good for facts about one brand, not for comparing |
- g2.com6×Missing
- pcmag.com5×Missing
- hubspot.com4×Missing
- mailchimp.com4×Missing
- zapier.com3×Missing
- techradar.com3×Missing
If a roundup lists five tools and Acme Email isn't one of them, an answer built from that roundup will usually name those five tools too. The AI isn't judging Acme Email. It simply never saw it.
Why your own site is rarely the source
Most brands assume their website is the main thing AI reads about them. For unbranded buyer questions, that's usually not true. Three reasons:
- Your site talks about you, not the market. A buyer question like “best email tool for a small shop” is a comparison. Search tends to return pages that compare, not one vendor's homepage.
- Your site is not neutral. Every brand says it is the best. A third-party page that compares five tools is more useful for answering a “which one” question.
- Your site wins different questions. When someone asks “What does Acme Email cost?” or “Is Acme Email good for Shopify?”, your own pages are a natural source. That's useful, but those buyers already know your name.
A note on quick fixes: Google's AI features guide (opens in a new tab) says you don't need new machine-readable files, AI text files or special schema markup to appear in its AI features. Files like llms.txt are cheap to add, but don't expect them to change who gets recommended.
What this means for you
If AI answers are built from a short list of pages, the work becomes concrete. You don't need to “optimize for AI” in the abstract. You need to know which pages AI reads for your buyers' questions, and whether you're on them.
- 1
Write down the questions your buyers ask
Use their words, not your keywords. For Acme Email: “best email tool for a small online shop”, “cheaper Mailchimp alternative”, “email tool with simple automation”.
- 2
See which pages the answers come from
Ask each question and note the source links. Ask more than once, because answers change between runs.
- 3
Find the gaps
List the pages that name your competitors but not you. These are your best targets.
- 4
Get listed, honestly
Ask customers for reviews, pitch roundup writers with real facts, complete directory profiles, and join forum threads where you can genuinely help. Our guide Get on the pages AI cites walks through each type.
Spotlit does steps 2 and 3 for you. It runs its own AI model with live web search, asks your buyer questions several times, and shows every page each answer was built from. It does not read ChatGPT or Perplexity directly, so treat it as a close look at the same kind of process, not a copy of one app's answer.
Common questions
Does AI always search the web before answering?
No. Some answers come only from what the model learned in training, and each tool decides when to search. For questions about products, prices and “best” lists, AI search tools often do search, which is why the pages they find matter.
Why does a competitor with a worse product get recommended?
Often because it's listed on more of the pages the AI reads: more reviews, more roundups, more forum mentions. The answer reflects what those pages say, not a fresh product test. That's good news: page presence is something you can work on.
If I'm on a page AI cites, will I be recommended?
Not guaranteed. Being on the page makes it possible, not certain. Answers vary between runs and between tools, so measure over several runs and over time.
Want to see which pages shape answers in your category? Scan your site free, then read how the Sources page works.
Found this useful? Share it with your team.
We build Spotlit and write these from how our own scans work, with every outside fact linked to its source.